In a delta modulated communication system, the analog signal to be transmitted is digitized into a stream of bits which subsequently may be reconstructed into a waveform approximating that of the original analog signal. In this wave reconstruction process, binary one bits cause the wave amplitude to increment in a positive sense while binary zero bits cause the wave amplitude to increment in a negative sense. The resulting stepped wave envelope follows the original wave envelope in a reasonably accurate manner. In a feedback loop at the transmitting end of the delta modulation system, the stepped wave envelope is constructed from the bits which are to be transmitted and this envelope is continuously compared with the original analog wave envelope to control the generation of the binary one or binary zero bits as may be required to construct the desired signal waveform at the system's receiver.
Since speech exhibits a wide range of signal levels and consists of talkspurts, a compandor circuit is required to preserve signal-to-noise ratio for all levels of signals and the fidelity of the signal. Delta modulation circuits incorporate various forms of adaptive companding techniques to satisfy voice applications. These adaptive companding techniques provide step size adjustments to the change of the input signal and enhance the dynamic companding range of the coding system. In general, adaptive compandors cause the step size to be reduced when the slope of the input signal is small and increase the step size when the slope is steep so as to provide better tracking over a wide range of input levels. However, prior art adaptive delta modulation systems have been confronted with the problems of quantization noise and slope overload noise. Quantization noise is introduced in the process of sampling the input analog waveform. Delta modulators must generate some quantization noise in order to produce the binary positive and negative voltage levels. In delta modulation, the rate of occurrence of each binary level is proportional to the slope of the analog input signal. When the slope increases or decreases fast enough to prevent the staircase feed back signal from tracking the input analog signal, a stream of binary ones or binary zeros will be output by the decoder. This condition is called "slope overload" which generally causes a much larger noise than does quantization noise, and which substantially reduces waveform fidelity. The slope overload further degrades the transient response.